databert 0.5.0

BERT Classifier for text classification

DataBERT Classifier for Dataset Identification

This package provides a comprehensive toolset for training and deploying BERT-based classifiers for text classification tasks, ranging from binary to multiclass problems. Leveraging the power of the transformers library, it simplifies the process of fine-tuning pre-trained BERT models and offers functionalities to evaluate, save, and deploy these models effectively.

Features

• Easy-to-use classes for training and using BERT models for text classification.
• Preprocessing and tokenization tailored for BERT's requirements.
• Calculation of various evaluation metrics including accuracy, precision, recall, specificity, and ROC-AUC.
• Functionality to save and load trained models and tokenizers.
• Capability to push trained models to the Hugging Face Hub.

Installation

To use this package, you need to install the required libraries. It's recommended to use a virtual environment:

pip install transformers torch sklearn numpy

Usage

Training DataBERT

1. Initialize the Classifier

You can initialize the TrainBERTClassifier with desired training parameters. Here's an example:

from bert_classifier import TrainBERTClassifier

classifier = TrainBERTClassifier(
    model_name='bert-base-uncased',
    num_labels=2,
    max_length=128,
    batch_size=32,
    learning_rate=2e-5,
    epochs=3
)

2. Prepare your data

Organize your text data and labels for training and validation. For example:

train_texts = ['This is the first text', 'Here is another one']
train_labels = [0, 1]

val_texts = ['This text is for validation', 'Another validation text']
val_labels = [0, 1]

3. Train the model

Use the train method to fine-tune the BERT model on your data:

classifier.train(train_texts, train_labels, val_texts, val_labels)

4. Evaluate the Model : After training, the model's performance metrics for the validation set will be printed automatically, including accuracy, precision, recall, specificity, and ROC-AUC.

5. Save the Model:

Save your trained model and tokenizer for later use:

classifier.save_model_and_tokenizer('path/to/save/directory')

Using the Trained BERT Classifier

After training and saving your model, you can use it for classifying new texts:

1. Initialize the Classifier with the Trained Model:

Load your trained model and tokenizer:

from bert_classifier import BERTClassifier

model_path = 'path/to/save/directory/model'
tokenizer_name = 'bert-base-uncased'  # Or path to tokenizer if you saved it

bert_classifier = BERTClassifier(model_path, tokenizer_name)

2. Classify new texts

You can now use the classifier to predict the class of new texts:

text = "Example text to classify"
prediction, confidence = bert_classifier.predict(text)
print(f"Predicted class: {prediction} with confidence {confidence}")

Contributions

Contributions to this package are welcome. Please follow conventional commit messages and ensure code quality for any pull requests.

License

This project is licensed under the MIT License - see the LICENSE file for details.